Spaces:
Running
Running
File size: 10,762 Bytes
fbfc907 ad2ab25 9f8226f 2978777 ad2ab25 fbfc907 2978777 d4efb04 ad2ab25 83e8d61 d4efb04 c3e7be2 83e8d61 5fbe3e5 c3e7be2 d4efb04 c3e7be2 83e8d61 5fbe3e5 c3e7be2 d4efb04 ad2ab25 2978777 b21e058 2978777 ad2ab25 f9d12de ad2ab25 fbfc907 ad2ab25 2978777 0179a13 2978777 ad2ab25 fbfc907 ad2ab25 fbfc907 ad2ab25 fbfc907 ad2ab25 589e0e8 a434986 ad2ab25 2978777 0179a13 2978777 b21e058 b52e67d 589e0e8 b52e67d 9f8226f 0aa6a75 589e0e8 0aa6a75 ad2ab25 0aa6a75 5a9d2e7 0aa6a75 589e0e8 5a9d2e7 ad2ab25 b21e058 ad2ab25 0179a13 6d6ce37 fbfc907 dd7d4a5 fbfc907 ad2ab25 dd7d4a5 ad2ab25 dd7d4a5 ad2ab25 dd7d4a5 ad2ab25 dd7d4a5 ad2ab25 589e0e8 dd7d4a5 589e0e8 dd7d4a5 ad2ab25 dd7d4a5 ad2ab25 dd7d4a5 ad2ab25 dd7d4a5 ad2ab25 fbfc907 589e0e8 b52e67d b21e058 ad2ab25 dd7d4a5 ad2ab25 0aa6a75 589e0e8 0aa6a75 ad2ab25 dd7d4a5 ad2ab25 02b5c61 589e0e8 0aa6a75 d10611d ad2ab25 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 |
import os
import gradio as gr
from gradio.components import Slider
import torch
from transformers import pipeline
import pandas as pd
# Model, information and examples ----------------------------------------------
MODEL_NAMES = ["FLOR-1.3B-GL","Cerebras-1.3B-GL"]
markdown_description_en = """
# Galician LLMs
This space contains the Galician language models developed by [Proxecto Nós](https://nos.gal/en/proxecto-nos).
💐 **[FLOR-1.3B-GL](https://huggingface.co/proxectonos/FLOR-1.3B-GL)** is a 1.3B parameters model which is a Continual pretraining from [FLOR-1.3B](https://huggingface.co/projecte-aina/FLOR-1.3B), which is based in [Bloom 1.7B](https://huggingface.co/bigscience/bloom-1b7).
👀 **Learn more about FLOR-1.3B-GL:** [HF official model card](https://huggingface.co/proxectonos/FLOR-1.3B-GL).
🧠 **[Cerebras-1.3B-GL](https://huggingface.co/proxectonos/Cerebras-1.3B-GL)** is a 1.3B parameters model based in [Cerebras-GPT 1.3B](https://huggingface.co/cerebras/Cerebras-GPT-1.3B).
👀 **Learn more about Cerebras-1.3B-GL:** [HF official model card](https://huggingface.co/proxectonos/Cerebras-1.3B-GL)
"""
markdown_description_gl = """
# LLMs de galego
Este espazo contén diferentes Grandes Modelos da Linguaxe feitos para o galego desenvolvidos polo [Proxecto Nós](https://nos.gal/en/proxecto-nos).
💐 **[FLOR-1.3B-GL](https://huggingface.co/proxectonos/FLOR-1.3B-GL)** é un modelo de parámetros 1.3B que é un preadestramento continuo de [FLOR-1.3B]( https://huggingface.co/projecte-aina/FLOR-1.3B), baseado a súa vez en [Bloom 1.7B](https://huggingface.co/bigscience/bloom-1b7).
👀 **Máis información sobre FLOR-1.3B-GL:** [tarxeta modelo oficial HF](https://huggingface.co/proxectonos/FLOR-1.3B-GL).
🧠 **[Cerebras-1.3B-GL](https://huggingface.co/proxectonos/Cerebras-1.3B-GL)** é un modelo de parámetros 1.3B baseado en [Cerebras-GPT 1.3B](https:/ /huggingface.co/cerebras/Cerebras-GPT-1.3B).
👀 **Máis información sobre Cerebras-1.3B-GL:** [tarxeta modelo oficial HF](https://huggingface.co/proxectonos/Cerebras-1.3B-GL)
"""
markdown_description ={"en": markdown_description_en,"gl": markdown_description_gl}
short_prompts_examples = [
["A receita tradicional das filloas é"],
["O neno vivía preto de"]
]
few_shot_prompts_examples = [
["Responde á seguinte pregunta. \nPregunta: \"Cal é a capital de Noruega? \"\nResposta: \"A capital de Noruega é Oslo.\"\n---- \nResponde á seguinte pregunta.\nPregunta: \"Cal é a moeda de Portugal\" \nResposta: \"A moeda de Portugal é o euro.\" \n---- \nResponde á seguinte pregunta. \nPregunta: \"Cal é a capital de Suecia?\"\nResposta:"],
["Extrae as entidades nomeadas do seguinte texto: \nTexto: \"Chámome Wolfgang e vivo en Berlin\" \nEntidades: Wolfgang:PER, Berlin:LOC \n ---- \nExtrae as entidades nomeadas do seguinte texto: \nTexto: \"María e Miguel non teñen ningún problema\" \nEntidades: María:PER, Miguel:PER \n---- \nExtrae as entidades nomeadas do seguinte texto: \nTexto: \"O mellor de Barcelona é o bar do meu amigo Pablo\" \nEntidades: Pablo:PER, Barcelona:LOC \n---- \nExtrae as entidades nomeadas do seguinte texto: \nTexto: \"Carlos comparte cuarto con Marc\" \nEntidades:"],
["Cualifica como Positivo ou Negativo o sentimento da seguinte frase:\n Texto: \"Estou moi feliz\"\n Polaridade: Positivo\n ---- \n Cualifica como Positivo ou Negativo o sentimento da seguinte frase:\n Texto: \"Non me gusta beber cervexa\"\n Polaridade: Negativo\n ----\n Cualifica como Positivo ou Negativo o sentimento da seguinte frase:\n Texto: \"O meu pai detesta o seu traballo\"\n Polaridade: Negativo\n ----\n Cualifica como Positivo ou Negativo o sentimento da seguinte frase:\n Texto: \"Uxía desfruta xogando ao fútbol\"\n Polaridade: Positivo\n ----\n Cualifica como Positivo ou Negativo o sentimento da seguinte frase:\n Texto: \"O neno non está contento coas notas\"\n Polaridade:"]
]
fronted_theme = 'Soft'
# Model charge ---------------------------------------------------------
model_id_flor = "proxectonos/FLOR-1.3B-GL"
generator_model_flor = pipeline("text-generation", model=model_id_flor)
model_id_cerebras = "proxectonos/Cerebras-1.3B-GL"
generator_model_cerebras = pipeline("text-generation", model=model_id_cerebras, token=os.environ['TOKEN_HF'])
# Load language texts ---------------------------------------------------------
df_interface = pd.read_csv("interface_texts.csv")
language = "gl"
# Generation functions ---------------------------------------------------------
def get_model(model_selection):
if model_selection == "FLOR-1.3B-GL":
return generator_model_flor
else:
return generator_model_cerebras
def remove_empty_lines(text):
lines = text.strip().split("\n")
non_empty_lines = [line for line in lines if line.strip()]
return "\n".join(non_empty_lines)
def predict(prompt, model_select, max_length, repetition_penalty, temperature):
print("Dentro da xeración...")
generator_model = get_model(model_select)
prompt_length = len(generator_model.tokenizer.encode(prompt))
generated_text = generator_model(
prompt,
max_length=prompt_length + max_length,
pad_token_id=generator_model.tokenizer.eos_token_id,
repetition_penalty=repetition_penalty,
temperature=temperature,
do_sample=True)
generated_sequence = generated_text[0]['generated_text']
if generated_sequence is None:
gr.Warning('Inference endpoint is not available right now. Please try again later.')
return
generated_sequence = remove_empty_lines(generated_sequence)
print("Xeración completada")
return generated_sequence
# Gradio app ---------------------------------------------------------
def get_text_lang(variable):
return df_interface.loc[df_interface['variable'] == variable, language].values[0]
def change_language(demo):
if language == "gl":
language = "en"
else:
language = "gl"
demo.launch()
def clear():
return (
None,
None,
gr.update(value=20),
gr.update(value=1.3),
gr.update(value=0.5)
)
def pass_to_input(generated_gl):
return (
gr.update(value=generated_gl),
None
)
def parameters_default(text):
return (
gr.update(value=30), # max_length
gr.update(value=1.3), # repetition_penalty
gr.update(value=0.5) # temperature
)
def parameters_fewshot_prompt(text):
return (
gr.update(value=15), # max_length
gr.update(value=1), # repetition_penalty
gr.update(value=0.5) # temperature
)
def gradio_app():
with gr.Blocks(theme=fronted_theme) as demo:
with gr.Row():
with gr.Column(scale=0.1):
change_lang = gr.Button(value=get_text_lang("change_lang"))
gr.HTML('<img src="https://huggingface.co/spaces/proxectonos/README/resolve/main/title-card.png" width="100%" style="border-radius: 0.75rem;">')
with gr.Column():
gr.Markdown(markdown_description[language])
with gr.Row(equal_height=True):
model_select = gr.Dropdown(
label=get_text_lang("model_select"),
choices=MODEL_NAMES,
value=MODEL_NAMES[0],
interactive=True
)
with gr.Row(equal_height=True):
with gr.Column():
text_gl = gr.Textbox(label=get_text_lang("text_gl"),
lines=6, placeholder="e.g. O neno vai a escola con ")
with gr.Row(variant="panel"):
with gr.Accordion(get_text_lang("accordion_parameters"), open=False):
max_length = Slider(
minimum=1,
maximum=200,
step=1,
value=30,
label=get_text_lang("max_length")
)
repetition_penalty = Slider(
minimum=0.1,
maximum=4,
step=0.1,
value=1.3,
label=get_text_lang("repetition_penalty")
)
temperature = Slider(
minimum=0,
maximum=1,
value=0.5,
label=get_text_lang("temperature")
)
generator_btn = gr.Button(value=get_text_lang("generator_btn"),variant='primary')
with gr.Column():
generated_gl = gr.Textbox(label=get_text_lang("generated_gl_label"),
lines=6,
placeholder=get_text_lang("generated_gl_placeholder"),
interactive=False,
show_copy_button=True)
pass_btn = gr.Button(value=get_text_lang("pass_btn"))
clean_btn = gr.Button(value=get_text_lang("clean_btn"))
generator_btn.click(predict, inputs=[text_gl, model_select, max_length, repetition_penalty, temperature], outputs=generated_gl, api_name="generate-flor-gl")
clean_btn.click(fn=clear, inputs=[], outputs=[text_gl, generated_gl, max_length, repetition_penalty, temperature], queue=False, api_name=False)
pass_btn.click(fn=pass_to_input, inputs=[generated_gl], outputs=[text_gl,generated_gl], queue=False, api_name=False)
change_lang.click(fn=change_language, inputs=[demo], outputs=[], queue=False, api_name=False)
with gr.Row():
with gr.Column(scale=0.5):
gr.Examples(
label = get_text_lang("examples_short_prompts"),
examples = short_prompts_examples,
inputs = [text_gl],
outputs = [max_length, repetition_penalty, temperature],
fn = parameters_default,
run_on_click = True
)
gr.Examples(
label = get_text_lang("examples_few_shot"),
examples = few_shot_prompts_examples,
inputs = [text_gl],
outputs = [max_length, repetition_penalty, temperature],
fn = parameters_fewshot_prompt,
run_on_click = True
)
demo.launch()
if __name__ == "__main__":
gradio_app() |